Brain Status

Brain Status

Display a quick dashboard of all projects in the Claude Brain graph — status, last activity, current focus, blockers.

Modes

• Dashboard (default, "brain status" / "show projects" / …): the per-project overview in "Dashboard Generation" below.
• Analytics ("brain stats" / "graph analytics" / "graph activity over time"): the aggregate counts in "Analytics (brain stats)" below.

Both resolve the graph path first (Prerequisites). Pick the mode from the trigger phrase; if ambiguous, default to Dashboard.

Prerequisites

Resolve the graph path per skills/_shared/path-resolution.md.

Dashboard Generation

1. Census first — this decides what appears. Glob (or ls) pages/Projects___*.md and pages/Tasks___*.md — one free call. This file list, not the ripgrep in step 2, is authoritative for "what's in my brain": a page belongs on the dashboard because it exists, not because a later grep happened to match its content. At census time you have filenames only, so apply only the exclusion a filename can decide: drop pages ending ___SessionArchive.md. Do not try to filter on type:: here — that requires reading each file, which would turn one free call into 45, and step 2's ripgrep already returns type:: for every page. The census is therefore a candidate list; step 2 narrows it.

2. Collect every census page's state in one call. Digest properties make the whole dashboard greppable:

   rg "^(type|status|last-updated|focus|next|open|digest-updated):: " pages/ \
      -g "Projects___*.md" -g "Tasks___*.md"
   

   One result set gives, per page with a readable property block: its type, status, freshness, current focus, next action, and any open blocker. This is where type:: becomes available, so this is where the type-based exclusions happen — the census (step 1) could not make them. Drop from the candidate list: any page with type:: session-archive, and any Projects___*.md whose type:: is not project — e.g. type:: task-index (a project's task inventory, like Projects___Unicorn-Globus___Tasks.md) or type:: project-note (a standalone note, like Projects___Unicorn-Globus___ClaudeCodeAutomation.md). Both match the glob but are not projects; counting them would inflate the dashboard and the "N projects have no digest" line with pages that were never supposed to carry one (skills/_shared/digest.md's scope rule). A census page with no type:: at all is not excluded here — it has a damaged or missing property block, and step 3 surfaces it rather than dropping it.

3. Reconcile the census against the grep. The grep is a content search, not the page list — a page whose property block is damaged or absent produces zero hits and would otherwise vanish from the dashboard entirely, which for a memory tool reads as "the project isn't there." Match every census file (step 1) against the hits (step 2); any census file with zero hits is listed explicitly — "ProjectName — no properties, run brain-doctor" — never silently dropped. The census, not the grep, decides what is listed.

4. Fall back per project, not wholesale. A project page that has some properties but no focus:: / next:: has not been backfilled with a digest yet (distinct from step 3's zero-hit pages, which have no readable properties at all). For those pages only, use the section-targeted reads in skills/_shared/section-locator.md — property block, first bullet of ## Current Plan, last entry of ## Session Log — exactly as before. A partially-backfilled graph therefore degrades one page at a time, never all at once. Mention the count once at the end: " projects have no digest — run brain-doctor to backfill." Report the real count; never a placeholder digit.

5. Apply staleness rules. Use skills/_shared/staleness.md to flag stale or abandoned projects. Also flag digest drift: step 2's results already carry both digest-updated:: and last-updated:: for every page with a digest, so the comparison costs no extra reads. Where the gap exceeds 30 days, the digest may be describing older content (e.g. another device still on v0.9.x saved without refreshing it, or a Logseq hand-edit) — flag it the same way stale projects are flagged, and mention the count once: " projects have a stale digest — see skills/_shared/digest.md to rebuild." Report the real count; never a placeholder digit.

6. Read cross-project decisions. Check pages/Decisions.md for entries from the last 30 days.

7. Read Meta date. Check pages/Meta.md last-updated:: only — don't read the whole file.

8. Task summary — no extra reads. The single ripgrep in step 2 already covered pages/Tasks___*.md. Group from those results: active and blocked tasks listed by ID with their status; done tasks collapsed to one count line ("N done"); tasks with no status:: listed as "legacy — run brain-doctor to backfill". Where a task has a focus::, show it; otherwise show the ID alone.

9. Present the dashboard. For each project: name, status, staleness annotation (if any), digest-drift annotation (if any), current focus, open questions/blockers. Then: task summary (from step 8), recent cross-project decisions, total counts.

10. Write a journey-log entry per skills/_shared/journey-log.md with activity line: viewed dashboard.

Example Output

Here's your brain status:

**LogseqBrain** (active, last updated <yyyy-MM-dd>)
Currently: <first bullet of Current Plan>
No blockers.

**ChivalricQuest** (active, last updated <yyyy-MM-dd>)
Currently: No active plan yet.
No blockers.

Tasks: CRMGM-2002 (active), CRMGM-1994 (blocked) · 31 done

2 projects tracked (2 active). No recent cross-project decisions.

Analytics (brain stats)

Read-only aggregate view. Writes nothing except the journey-log entry. Stay token-frugal — use skills/_shared/section-locator.md for targeted reads; never full-read project pages.

When counting, exclude template placeholder stubs — the italic markers a fresh brain-init page seeds, e.g. _Project-specific decisions._ under ## Decisions, _Session entries are added by brain-save._ under ## Session Log, and _No active plan yet._ under ## Current Plan. They denote an empty section, so a section that contains only its stub counts as 0, not 1.

1. Projects. Glob pages/Projects___*.md. Count total. Apply skills/_shared/staleness.md, then collapse its four levels into two buckets for the count: active = fresh + aging, stale = stale + abandoned. Exclude session-archive pages from the project count, and exclude pages whose type:: isn't project (task-index, project-note, etc. — same exclusion as Dashboard step 1) — they match the glob but aren't projects.

2. Decisions. Count two distinct figures, because cross-project decisions are intentionally duplicated in both places (so never sum them): (a) cross-project decisions in pages/Decisions.md, and (b) decisions recorded on project pages (in their ## Decisions sections; this includes the project-page copy of any cross-project decision). Break each down by status:: value (e.g. accepted, superseded).

3. Sessions. For each project page, count real entries under ## Session Log (section-targeted read; skip the placeholder stub). Sum across projects.

4. Activity (recent window). Glob journals/*.md. For journals dated within the last 30 days (filename yyyy_MM_dd.md), count bullets under ## Activity. Report the total as the recent activity signal.

5. Present a compact block. The Tasks line is computed as in Dashboard step 8 — grouped from the same single ripgrep as step 2, which already covers pages/Tasks___*.md; no per-task reads. legacy = task pages with no status::.

   Brain stats:
   
   Projects: <N> (<active> active, <stale> stale)
   Tasks: <N> (<a> active, <b> blocked, <d> done, <l> legacy)
   Decisions: <P> on project pages, <X> cross-project (by status: <accepted> accepted, <superseded> superseded)
   Sessions logged: <S>
   Activity (last 30 days): <A> entries
   

6. Write a journey-log entry per skills/_shared/journey-log.md with activity line: viewed brain stats.

Important Notes

• Concise overview, not a deep load — bias toward fewer reads.
• If no projects yet: "Your brain is empty. Use 'init brain project [name]' to add your first project."